Writing by algorithm

Welcome to AusStories 2.6

In just a few moments, you’ll be writing your very own Australian masterpiece with the help of the AusStories database, which contains over nine hundred million possible novels, short stories, essays and poems.

So begins Ryan O’Neill’s remix of a Cate Kennedy story for if:book’s ongoing project Lost in Track Changes. It’s a story that imagines the setup protocol for software that allows the user to create quality short stories at a whim (the $4.99 Winton upgrade guarantees a Miles Franklin nomination for example).

It’s a clever update to a premise that has been knocking around for some time: that we can imagine the creative act of writing being outmoded, superseded by machines or software. One my first short stories invented an artificial intelligence agent named ‘Hemmingway’ (the additional ‘m’ is never explained by the way) capable of producing any copy asked of him within moments.

Is it possible for our current wave of technological change to drive authorship into irrelevance?

Sure it’s possible. It’s happening right now to a certain extent. The Associated Press along with a host of top tech companies have already begun using Wordsmith, software capable of taking large and complex data sets such as stock quotes or sports results and turning them into readable stories. By all accounts, it works. And it’s uncanny.

But getting machines to spin prefab stories from boring stuff like stock market summaries is one thing. Let’s go to the other extreme. How would a machine go about recreating something like Patrick White’s prose? From an engineering point of view, we work with a finite number of symbols and mostly in broad patterns that can be predefined. Whether an algorithm can produce copy in the style of a sports summary or Nobel Prize-winning literature is just a case of using the right input, both in style and scale. With the right data and enough processing power, the next Voss might well be reproducible at the push of a button.

Not that there’s a great demand for another Voss. On a related point, I have a theory that Clive Cussler is already an algorithm.

But there’s one fundamental problem at the heart of this kind of innovation. We already know White’s style. We can analyse his body of work make inferences from it to construct writing in a way that’s reminiscent of how he arranged his words. Algorithms are great at recreating something known, analysing and reproducing according to a set of instructions.

Imagine if you were to sit down and re-write a story you know well exactly as it was originally authored. It could be anything: Heart of Darkness or The Da Vinci Code. Tell the story exactly as its original author did. You can’t, can you? No matter how hard you try, you’ll never get it the same. Ever tried to rewrite your own work from scratch after losing the original copy? Even when you were the original author, again, it never get quite comes out the same. Sometimes it’s worse. Sometimes it’s better.

What gets in the way of our ability to play photocopier? Memory, sure, but also voice. You can never quite escape your own voice, your own point of view, your own unique way of seeing and recording the world around you or the world in your head. It’s a part of you and only you that hits the page every time you do.

I’m not saying it’s impossible, but for now authors can sleep tight. The software still has a long way to go.


Lost in Track Changes
AP's automation software